Real-Time Crack Detection and Segmentation in Metallic Surfaces Using YOLO v11-Seg Model Integrated with HMI for Industrial Applications

Cracks in metallic surfaces fail industrial systems, compromising systems' safety and productivity. It is therefore imperative to detect such cracks in their initial stage, so preventive measures can be taken to avoid downtime and associated risks. In this paper, we report the novel adaptation and application of the YOLO v11-seg model for precise crack segmentation in metallic surfaces. The specialized configuration optimizes both detection accuracy and inference speed, making it suitable for industrial applications. The deep learning model is integrated into a Human-Machine Interface (HMI) using the Open Neural Network Exchange (ONNX) format, to enable seamless real-time visualization and interaction. This enhances usability for industrial operators, bridging the gap between advanced AI models and practical deployment. We report a fully functional and versatile inspection tool by combining the deep learning model with hardware and real-time video processing via Python. A custom dataset comprising 1,111 training and 246 test images was curated, annotated with segmentation masks, and augmented using the Albumentations library to improve generalization. The model showed detection and segmentation precisions of 96% and 94%, respectively.